SM011-13
Using Machine Learning and Geomagnetic Storm Data to Determine the Risk of GIC Occurrence

Tuesday, 8 December 2020: 07:36
Virtual
Michael Coughlan, University of New Hampshire, Physics, Durham, NH, United States, Amy M Keesee, University of New Hampshire, Physics and Space Science Center, Durham, NH, United States, Victor A Pinto, University of New Hampshire Main Campus, Institute for the Study of Earth, Oceans and Space, Durham, NH, United States, Jeremiah W Johnson, University of New Hampshire, Manchester, United States and Hyunju KIM Connor, University of Alaska Fairbanks, Fairbanks, AK, United States
Abstract:
Geomagnetically induced currents (GICs) can cause massive disruptions to electrical systems and other vital infrastructure. GIC events are more likely to occur during periods of geomagnetic disturbance and their magnitude is correlated to the intensity of the disturbance. In-situ GIC measurements are rarely available, so fluctuations in the horizontal component of the ground magnetic field are often used as a proxy for determining the risk of GIC occurrence. In this work, different machine learning techniques were investigated as a tool to predict the risk of GIC events by forecasting the horizontal component of dB/dt at several ground magnetometer stations at mid and high latitudes. Time dependent Feed Forward and Long-Short Term Memory (LSTM) Recurrent Neural Networks were used to model dBH/dt using OMNI solar wind data, and Supermag ground magnetometer data during geomagnetic storms that occurred between 1995-2019.
<< Previous Abstract | Next Abstract